Gazi Üniversitesi Fen Bilimleri Dergisi (Jun 2019)
The Probabilistic Prediction of Solar Energy Power Production Based on Time in Smart Grids
Abstract
Population growth and energy resource based on fossil fuel depletion increase the demand for renewable energy resources, especially for solar energy in the world. Smart grids have been developed in order to meet the growing energy need in the form of an intelligent structure with renewable energy sources. One key goal of the smart grid initiatives, therefore, increases the ratio of the renewable energy within overall energy power generation. However, the integration of renewable energies into the grid, whose power generation is intermittent and uncontrollable, leads to a number of challenges. It is critical to determine which renewable source will be dispatched to satisfy the variety of customer demands, and predict the energy power in advance. In this study, the energy generation could be modeled based on the weather measurements using the machine learning algorithms and the renewable energy production system oriented power generation could be, thus, predicted hourly. This model was created by machine learning approaches and an energy production estimate was made. A variety of methods such as multiple linear regression, Powell optimization and probabilistic programming based on Markov Chain Monte Carlo simulations were used and their capability of predictions were compared to each other. While energy production is estimated with an accuracy of 80% with an analytical approach, it has been predicted to be successful with a probabilistic approach, indicating the upper and lower limit of 95% confidence interval. These results show that the energy generation could be predictable based on the weather measurements using machine-learning algorithms. In addition, it is considered that estimation algorithms will facilitate the integration of renewable energy systems into the existing grid and make the smart grid more widespread.
